fix(pg-cursor): clear query_timeout when the cursor finishes

[DylanMerigaud]

Fixes #3497

When a client has query_timeout set, Client#query wraps the query's callback so the timer is cleared once the query finishes. pg-cursor never calls callback, so for a cursor the timer always ran to the end. When it fired, it called the callback with Query read timeout even though the cursor had already been read and closed and the client had ended. Until then the timer kept the cursor, its rows and the process alive, which is the retention reported in the issue.

With this change the cursor calls callback once: with its result when it ends, with the error when it fails. pg-query-stream already does the same on its cursor's end and error events. Code that passes no callback sees no difference; a callback passed to client.query(cursor, callback) is now called when the cursor finishes instead of only when the timeout fires.

Testing

• New packages/pg-cursor/test/query-timeout.js. One test reads and closes a cursor, the other makes it fail with SELECT 1 / 0. Both wait past a 100 ms query_timeout and check that the callback ran exactly once, without a timeout error. Both fail on master with Query read timeout and pass with this change.
• yarn lint passes.
• yarn lerna exec --concurrency 1 --ignore pg-native yarn test passes on Node 24.10 against PostgreSQL 17.6 set up like CI (SSL on, the SCRAM test roles). The pg-native suite crashes with SIGSEGV on my machine, on master too, so I left it out; this change does not touch it.

AI assistance

I drafted this with Claude Opus 5.5, read the diff line by line and can answer for every change.